Algorithms Have Reasons Behavior Ignores

The present contribution seeks to reconsider some of the fundamental stakes underlying machine-learning. The question of « behavior » will be problematized in light of Gilbert Simondon’s theory of the image cycle. By taking theoretical positions and technical ambitions in the field of machine-learning seriously, we suggest to frame the transformations of algorithmics in terms of two ideal-types: mechanical and behavioral. While taking care to avoid any binary opposition, we treat these two poles as successive structurations, phases of existence that transform the nature of the problem to be resolved, that operate a passage from one structure to another. Following Simondon, we qualify this passage as transductive so as to sketch an image of behavior that calls for a reinventing of our political and ethical frameworks. What if the question were: what collective is eventuated, or do we wish to eventuate, when we behave with algorithms?